BubbleWatch

Daily AI-bubble thermometer. Watches the AI equity complex, credit spreads, crowding and concentration signals; publishes a 0–100 Bubble Temperature, a regime classification and a self-scored direction forecast to GitHub Pages every morning.

Live site: https://tanishkbansode.github.io/bubblewatch/

What it does

1. Ingests daily: 22-name AI complex + power/utilities + benchmarks (Yahoo → Stooq fallback), FRED high-yield OAS & VIX (keyless CSV), quarterly hyperscaler capex / NVDA YoY (manual seed).
2. Computes six z-scored gauges → weighted Bubble Temperature → regime (expansion / euphoria / distribution / contraction / capitulation_recovery).
3. Trains a logistic P(AI complex up next 5d) on chronological splits, logs predictions to an append-only ledger, scores them against a permanent constant-0.50 baseline once horizons elapse.
4. Renders web/ (dashboard + three explainer pages), commits fresh data as bubblewatch-bot, deploys via GitHub Pages.

Local run

python -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt
python -m bubblewatch.pipeline run     # ingest + features + models + digest + web
python -m bubblewatch.pipeline report  # rebuild digest/site from existing lake
python -m bubblewatch.pipeline eval    # print prediction scoreboard

Outputs: logs/digest.md, web/index.html (+ explainer pages), lake parquet in data/lake/.

Quarterly maintenance

Append one row to data/manual/hyperscaler_capex.csv after each earnings season:

2025Q3,<combined MSFT+GOOGL+AMZN+META capex $B>,<NVDA revenue YoY %>

The dashboard surfaces staleness automatically.

Ops notes

• Schedule: daily 01:00 UTC (+ manual dispatch). The bot commits data logs web — always git pull --rebase before pushing local changes.
• Seed values through 2025Q2 are approximate from earnings releases; verify before trusting.

Disclaimers & Legal Notice

• Not investment advice: BubbleWatch is an educational and quantitative research project. It does not provide financial, investment, legal, tax, or accounting advice. All data and predictions are provided strictly "as is".
• Regulatory Disclosures & Terms: See the Legal Center for comprehensive regulatory disclosures, machine learning model limitations, terms of service, and zero-tracking privacy policy.
• License: Open source under the MIT License.